How do you convert a PDF into an Excel table?

Which method suits your PDF?

Three questions decide: does the PDF have a text layer, how many files are there, and may the contents leave your organisation? You can quickly see whether there is a text layer: if you can select and copy a word in the PDF, there is text. If not, the page is an image, usually a scan, and first needs text recognition (OCR).

None of the methods works flawlessly, and that is down to the PDF itself: according to Microsoft, it stores a table as a set of lines, with no relation to the content of the cells. Every tool has to infer the table from the position of the words. Typical pitfalls are multi-line cells, subtotals, repeated header rows and page breaks.

How does the PDF import in Excel work?

In Excel for Windows, you read in a PDF with Power Query, Excel's import tool. According to Microsoft's overview of data sources, the PDF import is part of Excel in a Microsoft 365 subscription (as of September 2026). It is missing from the perpetual versions Office 2016 and 2019, and Microsoft does not list PDF as a data source for Excel for Mac.

1. Start the import

   Choose Data → Get Data → From File → From PDF, select the file and click “Import”.

2. Select the table

   The Navigator shows the tables Excel has recognised in the PDF. Click on the entries and compare the preview with the PDF. If Excel does not find a usable table, first check whether the PDF has a text layer; a scan needs one of the methods from the next section.

3. Load, or clean up first

   “Load” puts the table straight into a new sheet. “Transform Data” first opens the Power Query Editor, where you set header rows, remove empty rows and rename columns.

4. Have numbers and dates read in German format

   Power Query reads numbers and dates according to your system's locale. If that does not match the PDF, right-click the column header in the editor, choose Change Type → Using Locale and set the data type and “German (Germany)”. That way, 1.234,56 and 31.08.2026 are recognised as a number and as a date.

5. Check tables that span pages

   By default, Power Query combines similar tables on consecutive pages. Still check whether header rows or carried-forward totals appear twice in the data.

6. Next time, just refresh

   If next month's PDF is in the same place under the same name, Data → Refresh All is enough. Excel then repeats all the clean-up steps.

If Excel reports on the first attempt that additional components are missing, the PDF import needs .NET Framework version 4.5 or later.

Can Copilot in Excel convert a PDF into a table?

Yes, with a Copilot licence and in a different way from the import: Copilot reads the PDF and builds the table itself instead of taking over a recognised table. Microsoft describes this in its Excel blog of 12 August 2026 in three steps: create a new workbook in Excel for the web and open the Copilot chat, attach the PDF using the plus icon and ask for it to be converted into a table, then check the table layout before applying formulas.

• Where it struggles, according to Microsoft: scanned or poorly legible PDFs, and password-protected, damaged or heavily formatted files. Remove password protection beforehand.
• What Copilot does differently from the import: it arranges rows, columns and headings at its own discretion. This helps with unusual layouts, but also means that you have to check every column against the PDF; a total check as in the checking section is part of this.
• Licence and data route: which Microsoft 365 licence Copilot in Excel requires, and when processing takes place outside the EU Data Boundary, is covered in the article Analyse Excel with AI.

What about scanned PDFs or a Mac?

A scan first needs text recognition, because it only contains an image of the page; the Excel import cannot read figures from it. On a Mac, the PDF import is missing altogether. Three methods can help, although the third only works with PDFs that have a text layer.

Are online converters a data protection risk?

Yes, as soon as the PDF contains personal or confidential data. By uploading, you hand the file over to an external provider; storage period, server location and access are then outside your control. Invoices, bank statements and payroll lists almost always contain names, bank details or amounts that can be attributed to a person.

If a service processes such data on your behalf, you need a data processing agreement under Art. 28 GDPR; what belongs in it is explained under DPAs for AI tools. For law firms, tax practices and medical practices, professional secrecy under § 203 StGB is added.

For files without personal data and without trade secrets, such as a public price list, an online converter is acceptable.

Our own measurement: what comes out of two test PDFs without rework?

On 24 September 2026, we ran two invented test PDFs with a text layer through four rule-based tools and counted how many table rows came out complete without any rework: an order with six items and table lines, and a two-page bank statement with 16 transactions and no lines, in which each transaction spans three lines. A row only counted as complete if all seven values of a record were in it as separate cells. We did not measure a language model, Excel, Copilot or Biwak.

• Lines and one row per record make the difference. The order came out complete with three of four methods, the bank statement with none.
• The bank statement was only complete with a rule for its layout: combine three lines into one transaction. This work is done by a person, a script or an AI agent; for a different bank format, the rule would have to be created anew.
• The balance check triggers. When we left out the last transaction, the calculation no longer added up. This check costs one formula and belongs in every conversion of a bank statement, whatever the tool.
• Business logic stays with you: no extractor recognises that “5 VE à 100” (five packs of 100) in the order means 500 pieces.

Many invoices, each with a different layout: when is an AI agent worthwhile?

When many PDFs have different structures and the end result should be a table with always the same columns, such as the line items on invoices from twenty suppliers. Excel's folder import requires the same structure, and writing a separate rule for each supplier, as in our measurement, is rarely worthwhile. An AI agent reads each file individually, assigns the values to your columns and can do the same check as we did: the total of the items against the invoice amount.

Biwak, a desktop app with an AI agent for office tasks, also works this way: the PDFs are in a working folder, and the result is created as a CSV file alongside them. Three limits apply in this case. Biwak reads scanned pages without a text layer only as images, without its own text recognition, and we promise no result for them; we do not promise finished .xlsx files for the desktop app; and the content used goes to the language model; for businesses, Biwak's data processing agreement applies. The data route is described on the security page, and use in accounting in the article AI in accounting. For bank statements there is a separate page with a sample statement: Convert bank statements to CSV and Excel. For a single clean table, the Excel import is faster.

How do you check the converted table?

Check every converted table at least for number of rows, totals, and date and number formats before you calculate with it. The list applies to each of the six methods, including Copilot and AI agents that build a table themselves.

Open a CSV file via Data → From Text/CSV instead of double-clicking it. When you double-click, according to Microsoft's help on CSV import, Excel applies the current default settings to every column; in the import dialog, you choose the delimiter and data type yourself. How to analyse the finished table afterwards is described in Analyse Excel with AI.

Frequently asked questions